ML Lab 2.0

Laying the foundation for ML

Posted by Scott Speights    on February 8, 2019 in Dev tagged with Culture, Conference, Development, Data

ML Lab 2.0 and The ML Foundation

Our ML Lab machine learning event had it’s second release on Tuesday and, different from our first ML Lab event, we now know a little bit about machine learning. But, while we’re no longer bloody beginners, it’s also safe to say we are still in the process of becoming ML engineering professionals.

Nevertheless, we’ve taken some big steps over the last couple of months, and since it’s hard to put a timeline on learning ML and completing ML projects, this is impressive, and something that should really motivate us to to close the loop on mastering machine learning.

Our ML Teaching Moment

Nine months ago, it was already clear that we wanted to start in incorporating machine learning into our products, but after agreeing on a firm delivery date for an ML Product with our customers and experiencing our first failures, we realized that we couldn’t move forward as fast as we would like. What was missing? Well, actually, almost everything.

In this discovery phase, we found out ( the hard way ) that AI and machine learning is not about deciding how to implement a software solution but about building a completely different kind of solution. One good indicator is the fact that this discipline requires having a combination of three different skill sets that were previously performed by three separate occupations.

Combine this with the fact that competent people are hard to find, and management decided it was time to embark on a program to lay the foundation for ML by providing our engineers with basic skills, gathering and democratizing data, and allowing experimentation to gain experience.

Machine Learning Progress on Display

ML Lab 2.0 showcased many of these achievements, provided the rest of the group with a snapshot of where we stand, who is doing what, and gave us a chance to network and ask questions. Highlights of the last nine months are -

That the first class of 21 ML engineers is finished with their training are starting to take a look at ML for their products, our data lake is coalescing, and will add the first freely available data shortly, and we have plenty of room to experiment, which is the mostly why people came to ML Lab 2.0 in the first place: To check out the experiments. Here is the rundown.

Hans Lecker - Presented initial research on the machine learning word2vec embedding project - where in cooperation with the TU Munchen - he and Felix Jablonski extracted keyword vocabularies using TF/IDF and vectorized this vocabulary into word embeddings. This ML feauture is now being used to recommend content to advocate customers.

Björn Waide and Eike Hirsch - talked about what Smartsteuer had done and what they wish to do with machine learning - from their first attempts at creating a chatbot, to being able to make useful predictions about Smarsteuer customer tax returns, to evolving these predictions into a series of new products.

Conrad Kleinn - Showed us the entire decision process of how to create customer segmentations using unsupervised learning with the k-means algorithm.

Yiftach Levy - Made two presentations. The first, showing how to perform text classifaction on e-mails with NLP to help customer service respond faster and more effectively. The second, with Patric Dokter of HCP marketing, presenting how customer analytics teamed up with marketing to predict customer actions - like churning - and to predict which countermeasures - like churn intervention programs - should be taken.

Alex Filip, Vladmir Mijatovic, and Agron Fazliu - Shared their organizational network analysis research Part 1 and Part 2 on how they created a solution to find found influencers by using machine learning to analyze communication patterns in e-mail on a public-domain data set.

Immediate ML Plans

Finally, our CTO, Raul, shared his vision for the (near) furure of ML at Haufe Group -

  • Starting round two of the AI Foundations training with AWS is starting in March to bootstrap another 21 Haufe Group software engineers to upskill ten percent of our engineers for data science.
  • Taking an intelligent approach to building ML solutions, by learning more about machine learning with high-level services ( think Azure Knowledge or AWS Comprehend ), that keep getting more powerful, to get a baseline, and to proceed with deeper technologies like Sagemaker or Tensor flow as needed.
  • Gradually removing the barriers to placing data in the cloud ( because it’s safe )

Eyeing the Finish line - ML Products - Scaled Deployments - Made in Haufe Group

Because software engineering is our core competency and because there are so few people out there on the job market, it’s important for Haufe Group to master machine learning and create solutions for customers on our own. It’s been a cool feeling to be part of organizing this program and watching it develop and succeed.

The foundation is there, and now, it’s time to travel the last mile of this journey by innovating responsibly with machine learning solutions and scaling them to production. Hope you can make it to ML Lab 3.0 and check out the next iteration of ML at Haufe Group…